@cartanova/qgrid-cli 2.4.6 → 2.4.8

package/bundle/src/application/request-log-step/request-log-step.entity.json

@@ -17,6 +17,26 @@
     },
     { "name": "step_index", "type": "integer", "desc": "ai-sdk step number (0-based)" },
     { "name": "type", "type": "enum", "id": "RequestLogStepType", "desc": "step 타입" },
+    {
+      "name": "model_name",
+      "type": "string",
+      "length": 50,
+      "nullable": true,
+      "desc": "실제 serving 모델"
+    },
+    {
+      "name": "requested_model_name",
+      "type": "string",
+      "length": 50,
+      "nullable": true,
+      "desc": "요청 모델 (fallback 전)"
+    },
+    {
+      "name": "fallback_count",
+      "type": "integer",
+      "nullable": true,
+      "desc": "모델 fallback 횟수"
+    },
     { "name": "input_tokens", "type": "integer", "nullable": true, "desc": "입력 토큰 (generate)" },
     {
       "name": "output_tokens",
@@ -31,6 +51,26 @@
       "nullable": true,
       "desc": "캐시 생성 토큰"
     },
+    {
+      "name": "cache_creation_5m_tokens",
+      "type": "integer",
+      "nullable": true,
+      "desc": "5분 TTL 캐시 생성 토큰"
+    },
+    {
+      "name": "cache_creation_1h_tokens",
+      "type": "integer",
+      "nullable": true,
+      "desc": "1시간 TTL 캐시 생성 토큰"
+    },
+    { "name": "cost_usd", "type": "integer", "nullable": true, "desc": "step 비용 (micro-USD)" },
+    {
+      "name": "cost_source",
+      "type": "string",
+      "length": 20,
+      "nullable": true,
+      "desc": "비용 출처(provider/pricing_table)"
+    },
     { "name": "duration_ms", "type": "integer", "nullable": true, "desc": "소요시간(ms)" },
     { "name": "ttft_ms", "type": "integer", "nullable": true, "desc": "Gen TTFT(ms)" },
     {
@@ -95,10 +135,17 @@
       "created_at",
       "step_index",
       "type",
+      "model_name",
+      "requested_model_name",
+      "fallback_count",
       "input_tokens",
       "output_tokens",
       "cache_read_tokens",
       "cache_creation_tokens",
+      "cache_creation_5m_tokens",
+      "cache_creation_1h_tokens",
+      "cost_usd",
+      "cost_source",
       "duration_ms",
       "ttft_ms",
       "finish_reason",
@@ -117,10 +164,17 @@
       "created_at",
       "step_index",
       "type",
+      "model_name",
+      "requested_model_name",
+      "fallback_count",
       "input_tokens",
       "output_tokens",
       "cache_read_tokens",
       "cache_creation_tokens",
+      "cache_creation_5m_tokens",
+      "cache_creation_1h_tokens",
+      "cost_usd",
+      "cost_source",
       "duration_ms",
       "ttft_ms",
       "finish_reason",

package/bundle/src/application/request-log-step/request-log-step.types.ts

@@ -11,10 +11,17 @@ export const RequestLogStepSaveParams = RequestLogStepBaseSchema.partial({
   id: true,
   created_at: true,
   request_log_id: true,
+  model_name: true,
+  requested_model_name: true,
+  fallback_count: true,
   input_tokens: true,
   output_tokens: true,
   cache_read_tokens: true,
   cache_creation_tokens: true,
+  cache_creation_5m_tokens: true,
+  cache_creation_1h_tokens: true,
+  cost_usd: true,
+  cost_source: true,
   duration_ms: true,
   finish_reason: true,
   reasoning_text: true,

package/bundle/src/application/sonamu.generated.http

@@ -103,10 +103,17 @@ Content-Type: application/json
       "request_log_id": 0,
       "step_index": 0,
       "type": "generate",
+      "model_name": null,
+      "requested_model_name": null,
+      "fallback_count": null,
       "input_tokens": null,
       "output_tokens": null,
       "cache_read_tokens": null,
       "cache_creation_tokens": null,
+      "cache_creation_5m_tokens": null,
+      "cache_creation_1h_tokens": null,
+      "cost_usd": null,
+      "cost_source": null,
       "duration_ms": null,
       "ttft_ms": null,
       "finish_reason": null,
@@ -173,6 +180,8 @@ Content-Type: application/json
       "token_name": null,
       "project_name": null,
       "model_name": null,
+      "requested_model_name": null,
+      "fallback_count": null,
       "user_prompt": null,
       "system_prompt": null,
       "response": "RESPONSE",
@@ -180,9 +189,12 @@ Content-Type: application/json
       "output_tokens": 0,
       "cache_read_tokens": 0,
       "cache_creation_tokens": 0,
+      "cache_creation_5m_tokens": null,
+      "cache_creation_1h_tokens": null,
       "duration_ms": 0,
       "ttft_ms": 0,
       "cost_usd": null,
+      "cost_source": null,
       "image_cost_usd": null,
       "image_cost_method": null,
       "effort": null,
@@ -360,10 +372,17 @@ Content-Type: application/json
     "requestLogId": 0,
     "stepIndex": 0,
     "type": "generate",
+    "modelName": "MODELNAME",
+    "requestedModelName": "REQUESTEDMODELNAME",
+    "fallbackCount": 0,
     "inputTokens": 0,
     "outputTokens": 0,
     "cacheReadTokens": 0,
     "cacheCreationTokens": 0,
+    "cacheCreation5mTokens": 0,
+    "cacheCreation1hTokens": 0,
+    "costUsd": 0,
+    "costSource": "provider",
     "durationMs": 0,
     "ttftMs": null,
     "finishReason": "FINISHREASON",
@@ -390,10 +409,17 @@ Content-Type: application/json
     "status": "succeeded",
     "response": "RESPONSE",
     "tokenName": "TOKENNAME",
+    "modelName": "MODELNAME",
+    "requestedModelName": "REQUESTEDMODELNAME",
+    "fallbackCount": 0,
     "totalInputTokens": 0,
     "totalOutputTokens": 0,
     "totalCacheReadTokens": 0,
     "totalCacheCreationTokens": 0,
+    "totalCacheCreation5mTokens": 0,
+    "totalCacheCreation1hTokens": 0,
+    "costUsd": 0,
+    "costSource": "provider",
     "totalDurationMs": 0,
     "history": "HISTORY",
     "errorMessage": "ERRORMESSAGE"

package/bundle/src/application/sonamu.generated.sso.ts

@@ -23,6 +23,8 @@ export const requestLogSubsetQueries = {
       token_name: "request_logs.token_name",
       project_name: "request_logs.project_name",
       model_name: "request_logs.model_name",
+      requested_model_name: "request_logs.requested_model_name",
+      fallback_count: "request_logs.fallback_count",
       user_prompt: "request_logs.user_prompt",
       system_prompt: "request_logs.system_prompt",
       response: "request_logs.response",
@@ -30,9 +32,12 @@ export const requestLogSubsetQueries = {
       output_tokens: "request_logs.output_tokens",
       cache_read_tokens: "request_logs.cache_read_tokens",
       cache_creation_tokens: "request_logs.cache_creation_tokens",
+      cache_creation_5m_tokens: "request_logs.cache_creation_5m_tokens",
+      cache_creation_1h_tokens: "request_logs.cache_creation_1h_tokens",
       duration_ms: "request_logs.duration_ms",
       ttft_ms: "request_logs.ttft_ms",
       cost_usd: "request_logs.cost_usd",
+      cost_source: "request_logs.cost_source",
       image_cost_usd: "request_logs.image_cost_usd",
       image_cost_method: "request_logs.image_cost_method",
       effort: "request_logs.effort",
@@ -50,13 +55,18 @@ export const requestLogSubsetQueries = {
       token_name: "request_logs.token_name",
       project_name: "request_logs.project_name",
       model_name: "request_logs.model_name",
+      requested_model_name: "request_logs.requested_model_name",
+      fallback_count: "request_logs.fallback_count",
       input_tokens: "request_logs.input_tokens",
       output_tokens: "request_logs.output_tokens",
       cache_read_tokens: "request_logs.cache_read_tokens",
       cache_creation_tokens: "request_logs.cache_creation_tokens",
+      cache_creation_5m_tokens: "request_logs.cache_creation_5m_tokens",
+      cache_creation_1h_tokens: "request_logs.cache_creation_1h_tokens",
       duration_ms: "request_logs.duration_ms",
       ttft_ms: "request_logs.ttft_ms",
       cost_usd: "request_logs.cost_usd",
+      cost_source: "request_logs.cost_source",
       image_cost_usd: "request_logs.image_cost_usd",
       image_cost_method: "request_logs.image_cost_method",
       effort: "request_logs.effort",
@@ -81,10 +91,17 @@ export const requestLogStepSubsetQueries = {
       created_at: "request_log_steps.created_at",
       step_index: "request_log_steps.step_index",
       type: "request_log_steps.type",
+      model_name: "request_log_steps.model_name",
+      requested_model_name: "request_log_steps.requested_model_name",
+      fallback_count: "request_log_steps.fallback_count",
       input_tokens: "request_log_steps.input_tokens",
       output_tokens: "request_log_steps.output_tokens",
       cache_read_tokens: "request_log_steps.cache_read_tokens",
       cache_creation_tokens: "request_log_steps.cache_creation_tokens",
+      cache_creation_5m_tokens: "request_log_steps.cache_creation_5m_tokens",
+      cache_creation_1h_tokens: "request_log_steps.cache_creation_1h_tokens",
+      cost_usd: "request_log_steps.cost_usd",
+      cost_source: "request_log_steps.cost_source",
       duration_ms: "request_log_steps.duration_ms",
       ttft_ms: "request_log_steps.ttft_ms",
       finish_reason: "request_log_steps.finish_reason",
@@ -105,10 +122,17 @@ export const requestLogStepSubsetQueries = {
       created_at: "request_log_steps.created_at",
       step_index: "request_log_steps.step_index",
       type: "request_log_steps.type",
+      model_name: "request_log_steps.model_name",
+      requested_model_name: "request_log_steps.requested_model_name",
+      fallback_count: "request_log_steps.fallback_count",
       input_tokens: "request_log_steps.input_tokens",
       output_tokens: "request_log_steps.output_tokens",
       cache_read_tokens: "request_log_steps.cache_read_tokens",
       cache_creation_tokens: "request_log_steps.cache_creation_tokens",
+      cache_creation_5m_tokens: "request_log_steps.cache_creation_5m_tokens",
+      cache_creation_1h_tokens: "request_log_steps.cache_creation_1h_tokens",
+      cost_usd: "request_log_steps.cost_usd",
+      cost_source: "request_log_steps.cost_source",
       duration_ms: "request_log_steps.duration_ms",
       ttft_ms: "request_log_steps.ttft_ms",
       finish_reason: "request_log_steps.finish_reason",

package/bundle/src/application/sonamu.generated.ts

@@ -92,6 +92,8 @@ export const RequestLogBaseSchema = z.object({
   token_name: z.string().max(100).nullable(),
   project_name: z.string().max(50).nullable(),
   model_name: z.string().max(50).nullable(),
+  requested_model_name: z.string().max(50).nullable(),
+  fallback_count: z.int().nullable(),
   user_prompt: z.string().nullable(),
   system_prompt: z.string().nullable(),
   response: z.string(),
@@ -99,9 +101,12 @@ export const RequestLogBaseSchema = z.object({
   output_tokens: z.int(),
   cache_read_tokens: z.int(),
   cache_creation_tokens: z.int(),
+  cache_creation_5m_tokens: z.int().nullable(),
+  cache_creation_1h_tokens: z.int().nullable(),
   duration_ms: z.int(),
   ttft_ms: z.int(),
   cost_usd: z.int().nullable(),
+  cost_source: z.string().max(20).nullable(),
   image_cost_usd: z.int().nullable(),
   image_cost_method: z.string().max(100).nullable(),
   effort: z.string().max(10).nullable(),
@@ -117,6 +122,8 @@ export type RequestLogBaseSchema = z.infer<typeof RequestLogBaseSchema> & {
     "token_name",
     "project_name",
     "model_name",
+    "requested_model_name",
+    "fallback_count",
     "user_prompt",
     "system_prompt",
     "response",
@@ -124,9 +131,12 @@ export type RequestLogBaseSchema = z.infer<typeof RequestLogBaseSchema> & {
     "output_tokens",
     "cache_read_tokens",
     "cache_creation_tokens",
+    "cache_creation_5m_tokens",
+    "cache_creation_1h_tokens",
     "duration_ms",
     "ttft_ms",
     "cost_usd",
+    "cost_source",
     "image_cost_usd",
     "image_cost_method",
     "effort",
@@ -146,10 +156,17 @@ export const RequestLogStepBaseSchema = z.object({
   request_log_id: z.int(),
   step_index: z.int(),
   type: RequestLogStepType,
+  model_name: z.string().max(50).nullable(),
+  requested_model_name: z.string().max(50).nullable(),
+  fallback_count: z.int().nullable(),
   input_tokens: z.int().nullable(),
   output_tokens: z.int().nullable(),
   cache_read_tokens: z.int().nullable(),
   cache_creation_tokens: z.int().nullable(),
+  cache_creation_5m_tokens: z.int().nullable(),
+  cache_creation_1h_tokens: z.int().nullable(),
+  cost_usd: z.int().nullable(),
+  cost_source: z.string().max(20).nullable(),
   duration_ms: z.int().nullable(),
   ttft_ms: z.int().nullable(),
   finish_reason: z.string().max(20).nullable(),
@@ -166,10 +183,17 @@ export const RequestLogStepBaseSchema = z.object({
 export type RequestLogStepBaseSchema = z.infer<typeof RequestLogStepBaseSchema> & {
   readonly __hasDefault__: readonly [
     "created_at",
+    "model_name",
+    "requested_model_name",
+    "fallback_count",
     "input_tokens",
     "output_tokens",
     "cache_read_tokens",
     "cache_creation_tokens",
+    "cache_creation_5m_tokens",
+    "cache_creation_1h_tokens",
+    "cost_usd",
+    "cost_source",
     "duration_ms",
     "ttft_ms",
     "finish_reason",
@@ -265,6 +289,8 @@ export const RequestLogSubsetA = z.object({
   token_name: z.string().max(100).nullable(),
   project_name: z.string().max(50).nullable(),
   model_name: z.string().max(50).nullable(),
+  requested_model_name: z.string().max(50).nullable(),
+  fallback_count: z.int().nullable(),
   user_prompt: z.string().nullable(),
   system_prompt: z.string().nullable(),
   response: z.string(),
@@ -272,9 +298,12 @@ export const RequestLogSubsetA = z.object({
   output_tokens: z.int(),
   cache_read_tokens: z.int(),
   cache_creation_tokens: z.int(),
+  cache_creation_5m_tokens: z.int().nullable(),
+  cache_creation_1h_tokens: z.int().nullable(),
   duration_ms: z.int(),
   ttft_ms: z.int(),
   cost_usd: z.int().nullable(),
+  cost_source: z.string().max(20).nullable(),
   image_cost_usd: z.int().nullable(),
   image_cost_method: z.string().max(100).nullable(),
   effort: z.string().max(10).nullable(),
@@ -291,13 +320,18 @@ export const RequestLogSubsetP = z.object({
   token_name: z.string().max(100).nullable(),
   project_name: z.string().max(50).nullable(),
   model_name: z.string().max(50).nullable(),
+  requested_model_name: z.string().max(50).nullable(),
+  fallback_count: z.int().nullable(),
   input_tokens: z.int(),
   output_tokens: z.int(),
   cache_read_tokens: z.int(),
   cache_creation_tokens: z.int(),
+  cache_creation_5m_tokens: z.int().nullable(),
+  cache_creation_1h_tokens: z.int().nullable(),
   duration_ms: z.int(),
   ttft_ms: z.int(),
   cost_usd: z.int().nullable(),
+  cost_source: z.string().max(20).nullable(),
   image_cost_usd: z.int().nullable(),
   image_cost_method: z.string().max(100).nullable(),
   effort: z.string().max(10).nullable(),
@@ -319,10 +353,17 @@ export const RequestLogStepSubsetA = z.object({
   created_at: z.date(),
   step_index: z.int(),
   type: RequestLogStepType,
+  model_name: z.string().max(50).nullable(),
+  requested_model_name: z.string().max(50).nullable(),
+  fallback_count: z.int().nullable(),
   input_tokens: z.int().nullable(),
   output_tokens: z.int().nullable(),
   cache_read_tokens: z.int().nullable(),
   cache_creation_tokens: z.int().nullable(),
+  cache_creation_5m_tokens: z.int().nullable(),
+  cache_creation_1h_tokens: z.int().nullable(),
+  cost_usd: z.int().nullable(),
+  cost_source: z.string().max(20).nullable(),
   duration_ms: z.int().nullable(),
   ttft_ms: z.int().nullable(),
   finish_reason: z.string().max(20).nullable(),
@@ -342,10 +383,17 @@ export const RequestLogStepSubsetT = z.object({
   created_at: z.date(),
   step_index: z.int(),
   type: RequestLogStepType,
+  model_name: z.string().max(50).nullable(),
+  requested_model_name: z.string().max(50).nullable(),
+  fallback_count: z.int().nullable(),
   input_tokens: z.int().nullable(),
   output_tokens: z.int().nullable(),
   cache_read_tokens: z.int().nullable(),
   cache_creation_tokens: z.int().nullable(),
+  cache_creation_5m_tokens: z.int().nullable(),
+  cache_creation_1h_tokens: z.int().nullable(),
+  cost_usd: z.int().nullable(),
+  cost_source: z.string().max(20).nullable(),
   duration_ms: z.int().nullable(),
   ttft_ms: z.int().nullable(),
   finish_reason: z.string().max(20).nullable(),

package/bundle/src/i18n/sd.generated.ts

@@ -215,6 +215,8 @@ const entityLabels = {
   "entity.RequestLog.token_name": "토큰 이름",
   "entity.RequestLog.project_name": "프로젝트 이름",
   "entity.RequestLog.model_name": "모델 이름",
+  "entity.RequestLog.requested_model_name": "요청 모델 이름 (fallback 전)",
+  "entity.RequestLog.fallback_count": "모델 fallback 횟수",
   "entity.RequestLog.user_prompt": "사용자 프롬프트",
   "entity.RequestLog.system_prompt": "시스템 프롬프트",
   "entity.RequestLog.response": "llm 응답 전문",
@@ -222,9 +224,12 @@ const entityLabels = {
   "entity.RequestLog.output_tokens": "출력 토큰",
   "entity.RequestLog.cache_read_tokens": "캐시 읽기 토큰",
   "entity.RequestLog.cache_creation_tokens": "캐시 생성 토큰",
+  "entity.RequestLog.cache_creation_5m_tokens": "5분 TTL 캐시 생성 토큰",
+  "entity.RequestLog.cache_creation_1h_tokens": "1시간 TTL 캐시 생성 토큰",
   "entity.RequestLog.duration_ms": "응답 소요시간(ms)",
   "entity.RequestLog.ttft_ms": "first gen TTFT(ms)",
   "entity.RequestLog.cost_usd": "API 비용",
+  "entity.RequestLog.cost_source": "비용 출처(provider/pricing_table/mixed)",
   "entity.RequestLog.image_cost_usd": "이미지 생성 추정 비용",
   "entity.RequestLog.image_cost_method": "이미지 비용 추정 방식",
   "entity.RequestLog.effort": "effort level",
@@ -247,10 +252,17 @@ const entityLabels = {
   "entity.RequestLogStep.request_log": "부모 request log ID",
   "entity.RequestLogStep.step_index": "ai-sdk step number (0-based)",
   "entity.RequestLogStep.type": "step 타입",
+  "entity.RequestLogStep.model_name": "실제 serving 모델",
+  "entity.RequestLogStep.requested_model_name": "요청 모델 (fallback 전)",
+  "entity.RequestLogStep.fallback_count": "모델 fallback 횟수",
   "entity.RequestLogStep.input_tokens": "입력 토큰 (generate)",
   "entity.RequestLogStep.output_tokens": "출력 토큰 (generate)",
   "entity.RequestLogStep.cache_read_tokens": "캐시 읽기 토큰",
   "entity.RequestLogStep.cache_creation_tokens": "캐시 생성 토큰",
+  "entity.RequestLogStep.cache_creation_5m_tokens": "5분 TTL 캐시 생성 토큰",
+  "entity.RequestLogStep.cache_creation_1h_tokens": "1시간 TTL 캐시 생성 토큰",
+  "entity.RequestLogStep.cost_usd": "step 비용 (micro-USD)",
+  "entity.RequestLogStep.cost_source": "비용 출처(provider/pricing_table)",
   "entity.RequestLogStep.duration_ms": "소요시간(ms)",
   "entity.RequestLogStep.ttft_ms": "Gen TTFT(ms)",
   "entity.RequestLogStep.finish_reason": "완료 사유",

package/bundle/src/migrations/20260718180824_alter_request_logs_add5_alter1.ts

@@ -0,0 +1,25 @@
+import { type Knex } from "knex";
+
+export async function up(knex: Knex): Promise<void> {
+  await knex.schema.alterTable("request_logs", (table) => {
+    // add
+    table.integer("cache_creation_1h_tokens").nullable();
+    table.integer("cache_creation_5m_tokens").nullable();
+    table.string("cost_source", 20).nullable();
+    table.integer("fallback_count").nullable();
+    table.string("requested_model_name", 50).nullable();
+  });
+}
+
+export async function down(knex: Knex): Promise<void> {
+  await knex.schema.alterTable("request_logs", (table) => {
+    // rollback - add
+    table.dropColumns(
+      "cache_creation_1h_tokens",
+      "cache_creation_5m_tokens",
+      "cost_source",
+      "fallback_count",
+      "requested_model_name",
+    );
+  });
+}

package/bundle/src/migrations/20260718180825_alter_request_log_steps_add7_alter1.ts

@@ -0,0 +1,29 @@
+import { type Knex } from "knex";
+
+export async function up(knex: Knex): Promise<void> {
+  await knex.schema.alterTable("request_log_steps", (table) => {
+    // add
+    table.integer("cache_creation_1h_tokens").nullable();
+    table.integer("cache_creation_5m_tokens").nullable();
+    table.string("cost_source", 20).nullable();
+    table.integer("cost_usd").nullable();
+    table.integer("fallback_count").nullable();
+    table.string("model_name", 50).nullable();
+    table.string("requested_model_name", 50).nullable();
+  });
+}
+
+export async function down(knex: Knex): Promise<void> {
+  await knex.schema.alterTable("request_log_steps", (table) => {
+    // rollback - add
+    table.dropColumns(
+      "cache_creation_1h_tokens",
+      "cache_creation_5m_tokens",
+      "cost_source",
+      "cost_usd",
+      "fallback_count",
+      "model_name",
+      "requested_model_name",
+    );
+  });
+}

package/bundle/src/utils/providers/anthropic/anthropic-constants.test.ts

@@ -6,6 +6,7 @@ import {
   canonicalAnthropicModel,
   hasOneMillionSuffix,
   needsCli1mSuffix,
+  requiresAdaptiveThinking,
   supports1MContext,
 } from "./anthropic-constants";
 
@@ -16,6 +17,7 @@ describe("canonicalAnthropicModel", () => {
 
   it("provider prefix 를 제거한다", () => {
     expect(canonicalAnthropicModel("anthropic/claude-opus-4-8")).toBe("claude-opus-4-8");
+    expect(canonicalAnthropicModel("anthropic/claude-fable-5")).toBe("claude-fable-5");
   });
 
   it("[1m] suffix 를 base canonical 에서 제거한다", () => {
@@ -26,6 +28,8 @@ describe("canonicalAnthropicModel", () => {
 
 describe("Anthropic 1M context policy", () => {
   it("지원 모델은 실측 확인된 exact set 만 true", () => {
+    expect(supports1MContext("claude-fable-5")).toBe(true);
+    expect(supports1MContext("claude-sonnet-5")).toBe(true);
     expect(supports1MContext("claude-sonnet-4-6")).toBe(true);
     expect(supports1MContext("claude-opus-4-6")).toBe(true);
     expect(supports1MContext("claude-opus-4-8")).toBe(true);
@@ -37,6 +41,8 @@ describe("Anthropic 1M context policy", () => {
   });
 
   it("CLI suffix 필요 모델과 기본 1M 모델을 분리한다", () => {
+    expect(needsCli1mSuffix("claude-fable-5")).toBe(false);
+    expect(needsCli1mSuffix("claude-sonnet-5")).toBe(false);
     expect(needsCli1mSuffix("claude-sonnet-4-6")).toBe(true);
     expect(needsCli1mSuffix("claude-opus-4-6")).toBe(true);
     expect(needsCli1mSuffix("claude-opus-4-8")).toBe(false);
@@ -58,3 +64,12 @@ describe("Anthropic 1M context policy", () => {
     expect(() => assertSupportedOneMillionSuffix("anthropic/claude-sonnet-4-6[1m]")).not.toThrow();
   });
 });
+
+describe("Anthropic thinking policy", () => {
+  it("Fable 5 만 always-on adaptive thinking 으로 실행한다", () => {
+    expect(requiresAdaptiveThinking("claude-fable-5")).toBe(true);
+    expect(requiresAdaptiveThinking("anthropic/claude-fable-5")).toBe(true);
+    expect(requiresAdaptiveThinking("claude-sonnet-5")).toBe(false);
+    expect(requiresAdaptiveThinking("claude-opus-4-8")).toBe(false);
+  });
+});

package/bundle/src/utils/providers/anthropic/anthropic-constants.ts

@@ -42,15 +42,18 @@ export function hasOneMillionSuffix(model?: string): boolean {
   return model !== undefined && ONE_MILLION_SUFFIX_RE.test(model);
 }
 
-// 1M 지원 여부는 Claude Code 유출본의 prefix 판정이 아니라 qgrid 실측 기준 exact set 으로 고정한다.
-// opus-4-7 은 실측 전까지 미포함. 새 모델 추가 시 실제 claude CLI contextWindow 동작으로 확인한다.
+// 1M 지원 여부는 Claude Code 내부의 광범위한 prefix 판정이 아니라 qgrid에서 확인한 exact set 으로
+// 고정한다. opus-4-7 은 실측 전까지 미포함. 새 모델 추가 시 공식 사양과 실제 CLI 동작을 확인한다.
 const ONE_MILLION_CONTEXT_MODELS = new Set([
+  "claude-fable-5",
+  "claude-sonnet-5",
   "claude-sonnet-4-6",
   "claude-opus-4-6",
   "claude-opus-4-8",
 ]);
 
 const CLI_ONE_MILLION_SUFFIX_MODELS = new Set(["claude-sonnet-4-6", "claude-opus-4-6"]);
+const ALWAYS_ADAPTIVE_THINKING_MODELS = new Set(["claude-fable-5"]);
 
 // 불변 계약: suffix 가 필요한 모델은 반드시 1M 지원 모델의 부분집합이어야 한다. 역방향 불일치(suffix
 // 대상인데 지원 집합엔 없음)면 needsCli1mSuffix=true·supports1MContext=false 가 동시에 나서
@@ -70,6 +73,12 @@ export function needsCli1mSuffix(model?: string): boolean {
   return CLI_ONE_MILLION_SUFFIX_MODELS.has(canonicalAnthropicModel(model));
 }
 
+// Fable 5 는 adaptive thinking 이 유일한 thinking mode 이므로 disabled 로 실행할 수 없다.
+// 나머지 모델은 qgrid 의 기존 저비용/비추론 경로를 그대로 유지한다.
+export function requiresAdaptiveThinking(model?: string): boolean {
+  return ALWAYS_ADAPTIVE_THINKING_MODELS.has(canonicalAnthropicModel(model));
+}
+
 export function assertSupportedOneMillionSuffix(model?: string): void {
   if (!hasOneMillionSuffix(model)) return;
   const canonical = canonicalAnthropicModel(model);

package/bundle/src/utils/providers/anthropic/anthropic-dispatcher.ts

@@ -347,7 +347,8 @@ export class AnthropicDispatcher implements ProviderDispatcher {
             ? `${session.text.slice(0, 500)}…(${session.text.length} chars)`
             : session.text
           : `empty text, outputTokens=${session.usage.outputTokens}`;
-        throw new Error(`claude error (${token.name}) [${reason}]: ${body}`);
+        const refusal = session.refusal ? ` refusal=${JSON.stringify(session.refusal)}` : "";
+        throw new Error(`claude error (${token.name}) [${reason}]: ${body}${refusal}`);
       }
 
       return {
@@ -357,7 +358,11 @@ export class AnthropicDispatcher implements ProviderDispatcher {
         durationMs: session.durationMs,
         ttftMs: session.ttftMs,
         costUsd: session.costUsd,
-        model,
+        // Claude Code 는 Fable refusal 을 Opus 로 자동 fallback 할 수 있다. 요청 모델을
+        // 그대로 serving model 로 기록하면 비용/감사 로그가 모두 틀어지므로 둘을 구분한다.
+        model: session.servedModel ?? model,
+        requestedModel: model,
+        modelFallbacks: session.modelFallbacks,
         // systemHash 는 상위 issueConvContext 가 채운다(여기선 비운다).
         threadCoord: {
           workerId: makeAnthropicWorkerId(token.id),

package/bundle/src/utils/providers/anthropic/claude-session.test.ts

@@ -12,6 +12,7 @@ import {
   oneMillionEnv,
   runClaudeSession,
   structuredOutputRetriesEnv,
+  thinkingEnv,
   SYSTEM_PROMPT_ARGV_MAX_BYTES,
   withSessionLock,
 } from "./claude-session";
@@ -132,6 +133,15 @@ describe("buildClaudeArgs (멀티턴/격리/structured)", () => {
     expect(oneMillionEnv(false)).toEqual({ CLAUDE_CODE_DISABLE_1M_CONTEXT: "1" });
   });
 
+  it("thinkingEnv: Fable 은 always-on adaptive thinking 을 보존하고 나머지는 비활성화", () => {
+    expect(thinkingEnv("claude-fable-5")).toEqual({});
+    expect(thinkingEnv("anthropic/claude-fable-5")).toEqual({});
+    expect(thinkingEnv("claude-opus-4-8")).toEqual({
+      CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING: "1",
+      MAX_THINKING_TOKENS: "0",
+    });
+  });
+
   // SON-495: structured output retry 를 1 로 고정한다.
   it("structuredOutputRetriesEnv: 미설정/비정상 값은 1 로 기본 고정", () => {
     expect(structuredOutputRetriesEnv(undefined)).toEqual({ MAX_STRUCTURED_OUTPUT_RETRIES: "1" });
@@ -221,6 +231,22 @@ describe("buildClaudeArgs (멀티턴/격리/structured)", () => {
     expect(args[args.indexOf("--model") + 1]).toBe("claude-opus-4-8");
   });
 
+  it("Fable 5 는 항상 켜진 adaptive thinking 을 끄지 않고 effort 만 전달한다", () => {
+    const args = buildClaudeArgs({
+      model: "claude-fable-5",
+      sessionId: "u",
+      effort: "low",
+    });
+    expect(args[args.indexOf("--model") + 1]).toBe("claude-fable-5");
+    expect(args).not.toContain("--thinking");
+    expect(args[args.indexOf("--effort") + 1]).toBe("low");
+  });
+
+  it("기존 Anthropic 모델은 thinking disabled 계약을 유지한다", () => {
+    const args = buildClaudeArgs({ model: "claude-opus-4-8", sessionId: "u" });
+    expect(args[args.indexOf("--thinking") + 1]).toBe("disabled");
+  });
+
   it("non-structured(jsonSchema 없음): --allowed-tools 미부여 (P2-7)", () => {
     const args = buildClaudeArgs({ model: "m", sessionId: "u" });
     expect(args).not.toContain("--allowed-tools");